Differences in Blood Leukocyte Subpopulations in Schizophrenia: A Systematic Review and Meta-Analysis

Large epidemiological studies have shown that severe infections are significant risk factors for schizophrenia and affective disorders,^1,2 with bacterial infections posing higher risk than viral and other types.³ Meta-analyses consistently reported elevated proinflammatory cytokines in blood and cerebrospinal fluid of patients with schizophrenia,^4,5 suggesting infections could trigger or exacerbate mental illness through cytokine action in genetically predisposed individuals or those with brain developmental risk.⁶

Recently, we observed elevated neutrophil counts in acutely ill unmedicated individuals with schizophrenia and major depression compared with controls.^7,8 Additionally, increased monocyte counts were found in schizophrenia. A decline in neutrophils during psychopharmacotherapy correlated with improvement in Positive and Negative Syndrome Scale (PANSS) and Hamilton Depression 21 scores. Elevated neutrophil counts in acute schizophrenia and major depression imply that bacteria may act as disease triggers, as neutrophils are key responders to bacterial infections.⁹

Meta-analyses of differential white blood cell (WBC) counts in psychosis are limited. We found only 4 systematic reviews that were published before 2024 and included 8 to 24 studies.^10-13 Two of these assessed neutrophil-lymphocyte or monocyte-lymphocyte ratios,^11,13 but reports on relative measures may mask absolute cell count changes. Another of these studies focused on monocyte counts only.¹² A more recent meta-analysis analyzed studies on absolute routine blood cell counts with a maximum of 33 analyzed studies per cell type and, regarding lymphocyte counts, mainly focused on studies using immunophenotyping.¹⁴ Reports indicate that during initial disease onset (first episode), immune activation may play a more prominent role, aligning with findings that inflammatory blood cytokine levels are even more increased during the first episode of psychosis.¹⁵ Antipsychotic drugs appear to modulate immune cell counts and reduce inflammation,^8,16,17 potentially reflected by specific immune cell elevations in patients who are antipsychotic naive or antipsychotic free. Elevated immune activity, indicated by higher leukocyte counts, may also be linked to more severe psychotic symptoms,^8,18,19 supporting the hypothesis that immune cell counts correlate with symptom severity and normalize after antipsychotic treatment.

We conducted this meta-analysis to determine if our previously reported findings⁸ align with other studies. We aimed to surpass previous research by analyzing absolute WBC numbers, systematically contacting authors for missing raw data, and exploring disease- and treatment-related changes and the influence of potential confounding factors like age, sex, body mass index (BMI), and smoking. We compared acutely ill patient subgroups with and without antipsychotic treatment, patients with first-episode vs chronic schizophrenia, and healthy controls to assess if blood count changes occur only in specific disease stages or independently of treatment. Additionally, we included longitudinal studies to observe changes in differential blood counts during acute psychosis treatment. Our main goal was to provide robust evidence for or challenge the immune hypothesis of schizophrenia.

Building on emerging evidence, we hypothesized that specific WBC populations:

• Are increased in schizophrenia compared with healthy controls.^8,11-14

• Show greater increases in first-episode vs chronic schizophrenia.¹⁵

• Show greater increases in untreated vs treated patients.^8,16,17

• Are linked to disease severity.^8,18,19

• Normalize after treatment with antipsychotic medication.^8,16-19

This meta-analysis compared absolute blood leukocyte counts between individuals with schizophrenia and healthy controls, adhering to Preferred Reporting Items for Systematic Reviews and Meta-Analyses (PRISMA) and Meta-Analysis of Observational Studies in Epidemiology (MOOSE) reporting guidelines (eAppendix in Supplement 1) and the Cochrane Handbook. Figure 1 outlines the study selection process, covering identification, screening, eligibility, and inclusion. Two authors (L.D. and M.N.) independently assessed study eligibility and performed data extraction in duplicate, with discrepancies resolved by a third author (J.S.).

We systematically searched PubMed, Web of Science, Scopus, and Cochrane library databases for peer-reviewed articles. The Boolean search terms were (schizophreni* OR psychosis) AND (leukocyt* OR granulocyt* OR neutrophil* OR eosinophil* OR basophil* OR monocyt* OR lymphocyt*). No restrictions were placed on year or country of publication with the last search performed in January 2024. Articles in languages other than English or German were translated using the DeepL Translator. Also, we examined reference sections of retrieved publications and other meta-analyses to identify additional articles. After eliminating reviews and meta-analyses, we narrowed the results according to the following inclusion and exclusion criteria:

Studies were included if they (1) included individuals 18 years or older with schizophrenia or schizoaffective disorder based on DSM-III through DSM-5, International Statistical Classification of Diseases and Related Health Problems, Tenth Revision, or Experiential World Inventory or Research Diagnostic Criteria (late 1970s standardized criteria for mental disorders influencing DSM-III); (2) involved pairwise comparisons with a healthy control group for between-group or pairwise comparisons before and after antipsychotic treatment for longitudinal within-group analyses; and (3) assessed leukocyte subpopulation counts in human blood samples in vivo.

Exclusion criteria were: (1) duplicate articles; (2) reviews/meta-analyses; (3) animal studies; (4) postmortem studies; (5) studies on pediatric/adolescent psychiatric populations (to avoid confounding specific to early-onset schizophrenia); (6) schizophrenia not investigated; (7) articles lacking original data; (8) case reports; (9) cross-sectional studies without healthy controls (eg, comparisons between neuropsychiatric disorders); (10) case-control studies with disease controls (ie, participants with psychiatric or somatic illnesses); (11) studies not reporting cell counts; (12) studies assessing leukocyte counts using only in vitro methods; (13) studies not reporting leukocyte subpopulation counts for all groups; (14) studies with leukocyte subpopulation counts affected by comorbidities (eg, infections, immunological/autoimmune diseases) or nonantipsychotic (eg, immunosuppressant) therapies; (15) longitudinal studies using switching or combination therapies, (16) longitudinal studies on clozapine adverse effects in treatment-resistant schizophrenia (to avoid confounding by clozapine’s known adverse effect of neutropenia); and (17) longitudinal studies with more than 12 weeks of follow-up not focused on acute illness treatment (to avoid confounding by increased variability in adherence and clinical course) (Figure 1).

Extracted data included diagnostic status, possible confounding factors (sex, age, disease duration, age at onset, smoking), leukocyte counts, medication status (medicated, antipsychotic medication–naive or -free), and symptom/disease severity scores (PANSS, Scale for the Assessment of Positive Symptoms [SAPS], Scale for the Assessment of Negative Symptoms [SANS], Brief Psychiatric Rating Scale [BPRS] and Clinical Global Impression [CGI]) (eTable 10 in Supplement 3). Baseline symptom severity scores were analyzed for longitudinal studies. Subgroup analyses considered antipsychotic status (taking antipsychotic medication, antipsychotic-free for at least 2 weeks, antipsychotic-naive at blood sampling), disease state (first-episode vs chronic schizophrenia), study setting (clinical vs population based), and secondary clinical conditions. We considered studies clinical if they recruited participants in controlled settings (eg, hospitals), and population based if they used broader recruitment (eg, cities, multiple hospitals) mostly via advertising. Clinical conditions were defined as comorbidities, including chronic diseases (eg, diabetes, hypertension), acute conditions (eg, infections, injuries), mental disorders, or other medical states requiring intervention. When data were unavailable, we contacted authors. For articles about the same cohort, we selected the one with the largest sample to maximize statistical power.

We assessed study quality using the Newcastle-Ottawa Scale (NOS)²⁰ with separate scales for case-control and cohort studies in between-group and within-group meta-analyses. The NOS evaluates selection, comparability, exposure (for case-control studies), and outcome (for cohort studies). Items indicating high quality were marked with stars. Each category has a maximum star count (selection = 4, comparability = 2, exposure = 3, and outcome = 3), yielding overall scores of 0 to 9 stars.²¹ For case-control studies, we awarded 1 star for “nonresponse rate” and “representativeness of the cases” even if not explicitly stated. Quality scores for included studies ranged from 4 to 9 stars (eTable 1 in Supplement 1). An NOS score of 4 or higher (at least moderate quality) was considered sufficient for inclusion in this meta-analysis.

Publication bias originates from favored publication of positive results. This was assessed using funnel plots that plot single-study effect size against a parameter representing study size, as well as Egger and Orwin fail-safe N tests. The latter estimates how many negative studies would be needed to make results nonsignificant (P > .05).²²

Data analyses were performed using R (version 4.3.1; metafor package version 4.6.0; R Foundation) and SPSS Statistics (version 28; IBM). We measured effect sizes using Hedges adjusted g with 95% CIs to assess differences in leukocyte counts. Hedges g is preferred for low bias and adjustability for small samples, with effect sizes classed as small (0.2), moderate (0.5), or large (0.8).²⁰ Hedges g was calculated if 3 or more studies were available. For between-group analyses, positive effect sizes indicated higher leukocyte counts. For within-group longitudinal comparisons, negative effect sizes indicated lower counts after medication. To control for familywise type I error in the primary outcome parameters, we applied Bonferroni adjustments to the P values by multiplying raw P values by 5, corresponding to the 5 analyzed leukocyte subpopulations (neutrophils, eosinophils, basophils, monocytes, and lymphocytes). Statistical significance was defined as Bonferroni-adjusted P values less than .05.

We assessed study heterogeneity using Cochrane Q, which measures deviation of single effect sizes from overall effect size (P < .10).²³ We also applied the I² metric with 50% to 75% indicating moderate and greater than 75% high heterogeneity.²⁴ Random-effects models accounted for variability.²⁵

Meta-regression analyses used the unrestricted maximum likelihood random-effects model²⁵ to evaluate how age, BMI, percentage of males, smoker percentage, follow-up length, disease severity (PANSS/SAPS/SANS/BPRS/CGI scores), age at onset, disease duration, sample size, and NOS scores influenced effect sizes. Studies were weighted by precision (95% CI and sample size). These analyses were considered exploratory; no Bonferroni correction was applied.

Median and quartiles were converted to mean and standard deviation, assuming normal distribution. In studies with missing PANSS scores but reporting BPRS values, we applied the linear regression approach by Leucht et al²⁶ to estimate PANSS total scores. Similarly, SANS and SAPS values were converted to PANSS negative and PANSS positive scores using the regression model by van Erp et al.²⁷

The meta-analysis comprised (1) overall analysis as a basis for subsequent stages, (2) sensitivity evaluation via leave-1-out analysis (a sensitivity method excluding 1 study at a time to assess result robustness and identify studies disproportionately affecting effect size or heterogeneity), and (3) meta-regression to identify potential effect size moderators.

We included 64 publications in the meta-analysis: 60 cross-sectional/case-control studies and 4 longitudinal studies. The study selection process is shown in Figure 1.

The 64 studies provided 62 pairwise comparisons of schizophrenia vs controls for between-group meta-analysis^{19 ,28-86} and 2 longitudinal studies without controls^{87 ,88} (eTables 2 and 3 in Supplement 1). These studies, published from 1972 to 2024, included 26 349 individuals with schizophrenia and 16 379 healthy controls, with a mean age of 23 to 51 years. Forty-three publications^{28-30,32,33,35,38,40,42-44,47-49,51,52,54-56,58,59,61,62,65-68,70-73,75,76,78-81,83-88} focused on chronic schizophrenia, 18 on first-episode schizophrenia,^{19,34,36,37,39,41,45,46,50,53,57,60,63,64,69,74,77,82} and 3 publications^8,31,89 on both groups. Thirteen articles^{8,19,34,45-47,49,53,55,56,64,66,88} reported all major absolute leukocyte subpopulation counts, and 39 matched patients and controls for age and sex.^{8,19,28-30,33,34,38,41-43,45-47,49,51-57,59,64,66,69,71-76,78,80,81,83-86} BMI data were available in 23 articles,^{8,19,30,31,42,45,47,49,51,53-57,59,60,64,73,74,76,78,82,84} 12 included patients without antipsychotic treatment,^{28,31,33,36,37,45,50,57,62,63,74,75,77,82,86} and 42 exclusively examined patients taking medication.^{19,29,30,32,34,35,38-44,46-49,51,52,54-56,59-61,64-73,76,78-81,83-85} Three studies^53,58,86 analyzed mixed cohorts of antipsychotic-naive/-free and medicated patients. Four longitudinal studies examined patients who were antipsychotic-naive or -free at baseline and after treatment. Only 20 studies reported smoking data^{8,34,41,42,45,46,51,54,55,57,60,63,64,69,72-74,76,78,84} and 31 used automated cell count methods.^{8,19,28,29,31,33,36-38,41,43-45,48,49,51,54-56,58,59,64,66,68,72,78,82,84-86,88}

Random-effects meta-analysis found higher neutrophil and monocyte counts in schizophrenia compared with controls (Table 1, Table 2, Figure 2, and Figure 3). Effect sizes were moderate for neutrophils (g = 0.69; 95% CI, 0.49 to 0.89; Bonferroni-adjusted P < .001; n = 40 951 [47 between-group comparisons]) and monocytes (g = 0.49; 95% CI, 0.24 to 0.75; Bonferroni-adjusted P < .001; n = 40 513 [44 between-group comparisons]). Differences in eosinophils (g = 0.02; 95% CI, −0.16 to 0.20; Bonferroni-adjusted P > .99; n = 3277 [18 between-group comparisons]), basophils (g = 0.14; 95% CI, −0.06 to 0.34; Bonferroni-adjusted P = .85; n = 2614 [13 between-group comparisons]), and lymphocytes (g = −0.08; 95% CI, −0.21 to 0.06; Bonferroni-adjusted P > .99; n = 41 693 [59 between-group comparisons]) were not significant (eTables 4-6 and eFigures 1-4 in Supplement 1).

Neutrophil counts were higher with large effect sizes in first-episode schizophrenia compared with healthy controls (g = 0.85; 95% CI, 0.35-1.34; Bonferroni-adjusted P = .004; n = 2602 [17 between-group comparisons]) but moderately elevated in chronic schizophrenia compared with healthy controls (g = 0.61; 95% CI, 0.44-0.77; Bonferroni-adjusted P < .001; n = 38 349 [30 between-group comparisons]). Monocyte counts showed a large effect size in first-episode schizophrenia (g = 0.91; 95% CI, 0.25-1.57; Bonferroni-adjusted P = .03; n = 2308 [16 between-group comparisons]) but only small effects in chronic schizophrenia (g = 0.28; 95% CI, 0.13-0.44; Bonferroni-adjusted P = .002; n = 38 205 [28 between-group comparisons]) (Table 1 and Table 2). Eosinophil, basophil, and lymphocyte effect sizes for patients with first-episode and chronic schizophrenia compared with healthy controls were similar (eTables 4-6 in Supplement 1).

Neutrophil differences were large in patients who were antipsychotic naive or free (Table 1), whereas effect size was moderate in patients taking medication. Monocyte differences followed a similar pattern, with the largest effect in antipsychotic-naive patients (Table 2). (Although there were only 2 pairwise comparisons for monocyte counts in antipsychotic-free individuals, we included these data because of the large effect size and for completeness.) Studies excluding comorbidities^{8 ,28-32,35-39,41,43-66,68,70-86,88,89} showed a greater neutrophil and monocyte effect size (Table 1 and Table 2).

Between-group meta-regression analyses found no significant relationship between neutrophil counts and moderators like sex, BMI, age, smoking, duration of illness, age at onset, PANSS scores, sample size, and NOS scores (Table 1). Monocyte count was related to BMI in patients who were antipsychotic-naive or -free (Table 2).

Four longitudinal studies met inclusion criteria. Two of these^8,89 analyzed first-episode and chronic schizophrenia separately, leading to 6 pairwise comparisons of baseline vs follow-up data. Because of the limited number of studies, no separate moderator analyses were performed for longitudinal data to avoid overinterpretation.

Neutrophils decreased longitudinally (g = −0.30; 95% CI, −0.45 to −0.15; Bonferroni-adjusted P < .001; n = 896 [4 within-group comparisons]) and eosinophils increased longitudinally (g = 0.61; 95% CI, 0.52 to 0.71; Bonferroni-adjusted P < .001; n = 876 [3 within-group comparisons]) after successful treatment of acute psychosis (Table 1 and eTable 4 and eFigures 4 and 6 in Supplement 1). Table 2 and eTables 5 and 6 and eFigures 5, 7, and 8 in Supplement 1 contain data for monocytes, basophils, and lymphocytes.

Heterogeneity was explored by comparing leukocyte differences across 54 clinical^{8 ,28-36,38-40,42-44,47-56,58,59,61-63,65-67,69-83,85-89} and 10 population-based studies.^{19 ,37,41,45,46,57,60,64,68,84} Neutrophil effect sizes were moderate in both settings, with larger effects in clinical studies (Table 1). Monocyte effect size was moderate in clinical studies but not significant in population-based settings (Table 2). No such differences were found for eosinophils, basophils, and lymphocytes (eTables 4-6 in Supplement 1). Leave-1-out analysis confirmed significant overall neutrophil and monocyte effects, with heterogeneity remaining significant regardless of the study excluded (eTables 7-8 in Supplement 1).

Funnel plots and Egger tests showed no significant bias for monocyte, eosinophil, basophil, or lymphocyte counts, but significant publication bias was detected for neutrophil differences (eFigures 9-13 in Supplement 1). The Orwin fail-safe test indicated 12 710 studies would be needed to nullify the neutrophil findings. No bias analyses were conducted for longitudinal data because of the small number of studies.

In 51 of 62 studies,^{8 ,19,28-30,33-35,38,39,41-43,45-49,51-60,62-66,68-81,83,85,86,88} patients and controls were age-matched. Five studies^{31 ,32,40,50,82} were sex-matched only, and 6 were unmatched for age or sex.^{28 ,36,37,44,61,67,89} Effect sizes for age- and sex-matched studies were similar to overall leukocyte subpopulation effects (Table 1, Table 2, and eTables 4-6 in Supplement 1).

To our knowledge, this meta-analysis is the largest to date examining absolute WBC subpopulation counts in patients with schizophrenia and healthy controls, focusing on disease- and treatment-related differences and potential confounders. Neutrophil and monocyte levels were notably higher in first-episode schizophrenia compared with chronic cases, suggesting differences in immune cell counts at various disease stages. Furthermore, neutrophils and monocytes were largely elevated in patients who were antipsychotic-naive or -free but moderately elevated in those taking antipsychotics.

These findings from the cross-sectional comparisons align with our previous research⁸ showing higher neutrophil and monocyte counts in first-episode vs chronic schizophrenia, but contrast with findings from Jackson et al.¹⁰ Increased neutrophil and monocyte counts support the immune hypothesis of schizophrenia, emphasizing innate immune activation. These cells produce proinflammatory cytokines, which are elevated in acute schizophrenia.⁴ Elevated neutrophils, key in bacterial defense,¹⁰ could suggest infections trigger acute psychosis in vulnerable individuals.³ Altered immune cell counts might also stem from congenital immune changes, supported by genetic studies linking leukocyte counts to schizophrenia.⁹⁰

Larger leukocyte differences in patients who were antipsychotic-naive or -free compared with patients receiving medication may result from neutropenic and monocytopenic effects of antipsychotics like clozapine and olanzapine, which reduce inflammation and improve symptoms.^91-96 Alternatively, antipsychotic medication may dampen the immune system by reducing inflammatory cytokines.^97,98 These findings, together with the current literature, support the hypothesis that immune cell counts and inflammation are higher in patients who were antipsychotic-naive or -free than in treated individuals with schizophrenia.

The more prominent findings in patients with first-episode vs relapsed disease suggest an early peak of immune activation that may trigger disease onset and progression. This aligns with inflammatory blood markers being associated with transition to psychosis in individuals at high clinical risk.^99,100 However, the persistent neutrophil elevation in schizophrenia suggests a broader role of chronic low-grade inflammation, which probably also contributes to the increased cardiovascular and metabolic risk linked to this disease.¹⁰¹ Inflammation may underlie both metabolic dysfunction and schizophrenia. Our finding that BMI is associated with monocyte counts in patients who were antipsychotic naive or free aligns with obesity-related low-grade inflammation,^102,103 highlighting the need to consider BMI and metabolic syndrome as covariates in future studies. Importantly, our results show that elevated neutrophils and monocytes are not solely explained by smoking, suggesting an intrinsic link to schizophrenia itself. These findings highlight the need for longitudinal studies to investigate immune mechanisms across disease stages, considering both systemic and brain-specific effects of chronic inflammation.

Longitudinal comparisons also support a decrease in neutrophil counts and an increase in eosinophil counts after antipsychotic treatment, possibly due to normalization of the inflammatory state, symptom improvement, or immunomodulating medication effects. No associations between longitudinal neutrophil differences and disease modifiers were found, likely because of the limited number of studies.

In contrast to previous studies linking neutrophil^7,8 and monocyte counts^104,105 with disease severity, our meta-analysis found no such associations except for eosinophils. This discrepancy may stem from variations in the number and characteristics of studies that reported PANSS scores, which could influence overall findings.

Clinical studies typically show larger effect sizes than population-based studies because of smaller, more homogeneous groups with severe symptoms and controlled settings.¹⁰⁶ We confirmed this for neutrophils but not for monocytes. The significant neutrophil and monocyte increases in population-based studies suggest our findings apply broadly, despite potential confounding variables.

Leave-1-out analyses confirmed the robustness of our results, with effect sizes unaffected by any single study. The observed heterogeneity likely reflects individual differences in WBC counts influenced by genetic, lifestyle, and environmental factors, as well as study-specific variables like demographics, antipsychotic treatment status, and disease stage. While subgroup analyses mitigated some heterogeneity, persistent variability underscores the complex relationship between immune responses and schizophrenia. This heterogeneity adds real-world relevance. It highlights the variability of immune dysregulation in schizophrenia.

The main strength of this meta-analysis is its inclusion of the largest number of studies to date on absolute leukocyte subpopulation counts (64 studies, 26 349 patients with schizophrenia, 16 379 healthy controls), providing high statistical power and enabling subgroup analyses to identify effects of disease state and medication (eTable 9 in Supplement 1). With a wealth of data from contacting original study authors and its longitudinal component, our meta-analysis expands and solidifies knowledge in this key area. We assessed disease severity using PANSS criteria, whereas previous studies relied on inpatient vs outpatient status or did not address this point.^{10 ,14} We also conducted meta-regression to explore associations with various potential moderators (eTable 9 in Supplement 1).

One limitation is that subgroup sample size imbalances and missing BMI and PANSS data in some studies reduced the certainty of meta-regression results. Other moderators such as race and ethnicity, genetic and epigenetic data, substance use, and environmental stressors were not included because of insufficient data. Sensitivity analyses suggested that heterogeneity could not be explained by excluding specific studies. Potential publication bias was indicated for neutrophil counts, but the Orwin fail-safe test suggested minimal bias. Lastly, effects of specific antipsychotics could not be determined because of a lack of detailed data in most studies.

This meta-analysis revealed elevated neutrophil and monocyte counts in schizophrenia, particularly in individuals with first-episode schizophrenia and those who were antipsychotic-naive or -free. The observed decrease in neutrophil counts and increase in eosinophil counts after antipsychotic treatment may reflect treatment-related immune modulation rather than acute illness itself. Notably, the longitudinal increase in eosinophils is consistent with the concept of the “eosinophil-driven dawn of recovery,” a phenomenon observed during the resolution of bacterial inflammation.¹⁰⁷ However, the significance of this finding in schizophrenia remains unclear due to the lack of data on microbial exposure in the original studies analyzed in our meta-analysis.

Our study underscores the importance of considering potential disease modifiers and underlying conditions, such as BMI, when examining immune-related changes in schizophrenia, paving the way for new therapeutic approaches grounded in the etiopathology of the disease.